39-Dehbi

8/13/2019 39-Dehbi

1/18

ASSESSMENT OF A NEW FLUENT MODEL FOR PARTICLE

DISPERSION IN TURBULENT FLOWS

A. Dehbi

Paul Scherrer Institut, Department of Nuclear Energy and Safety,

Laboratory forThermal-hydraulics

5232 Villigen, Switzerland

Abstract

In Lagrangian particle dispersion modeling, the assumption that turbulence is isotropic everywhereyields erroneous predictions of particle deposition rates on walls, even in simple geometries. In this

investigation, the stochastic particle tracking model in FLUENT 6.2 is modified to include a better

treatment of particle-turbulence interactions close to walls where anisotropic effects are significant.

The fluid rms velocities in the boundary layer are computed using fits of DNS data obtained in

channel flow. The new model is tested against correlations for particle removal rates in turbulent pipe

flow and 90obends. Comparison with experimental data is much better than with the default model.

The model is also assessed against data of particle removal in the human mouth-throat geometry where

the flow is decidedly three dimensional. Here, the agreement with the data is reasonable, especially in

view of the fact that the DNS fits used are those of channel flows, for lack of better alternatives. The

CFD Best Practice Guidelines are followed to a large extent, in particular by using multiple grid

resolutions and at least second order discretization schemes.

703

8/13/2019 39-Dehbi

2/18

1. Introduction

Particle-laden turbulent flows are common place in many and industrial, environmental and

medical applications. Examples include nuclear fission products transport, drug delivery in human

airways, fouling in compressor and turbine blades, pollutant control, etc. Therefore, an accurate

description of particle transport and removal in turbulent flow fields is of great importance. Most of

the investigations of particle-turbulence interactions have concentrated on unbounded flows whereboundary layer wall effects are nonexistent. Comparatively, wall bounded flows have attracted much

less attention. To a large extent, the deposition of particles onto the walls in turbulent flows is dictated

by what happens in the boundary layer. Indeed, in many applications, turbulence can be approximated

as being isotropic in the bulk; however, near the wall, it is known that the turbulent velocity

fluctuations are not isotropic, and the rms of the normal component of velocity can be orders of

magnitudes smaller than the streamwise or spanwise components. Hence, particles in the turbulent

boundary layer are slowed considerably compared to what would be expected if the flow field were

wholly isotropic.

Owing to the ever-increasing power of computer hardware, complex turbulent flows in nuclear

safety applications are increasingly predicted using Computational Fluid Dynamics (CFD) tools.

Given that fission product transport in reactor circuits is a major safety issue, it is only fitting that the

behavior of particles also be addressed within the CFD methodology. In widely used CFD codes such

as FLUENT [1], the turbulent dispersion of particles is predicted using a Lagrangian-type random

walk model, while assuming turbulence is isotropic all the way to the wall. The latter assumption leads

to large over-predictions of particle deposition rates even for simple flows such as those in pipes or

parallel channels. In the following investigation, a model is presented whereby the default random

walk model in FLUENT is modified to account for the anisotropic turbulence effects in the boundary

layer.

2. Particle transport methods

There are two main families of methods to treat particle transport in fluid flows: Eulerian and

Lagrangian. In the Eulerean or two-fluid approach, the particles are regarded as a continuous phasefor which the conservation equations (continuity, momentum and energy) are solved in similar fashion

to the carrier gas flow field. The Eulerean approach is particularly suitable for denser suspensions

when particle-particle interactions are important and the particle feedback on the flow is too large to

ignore. The main challenge facing Eulerian-type, two-fluid approaches resides in accurately defining

the inter-phase exchange rates and closure laws which arise from the averaging procedures. In addition,

the strong coupling between the phases renders the Eulerean approach quite delicate to handle,

especially at boundaries where the solid phase may be removed.

In the second approach, called Lagrangian, the particles are treated as a discrete phase made of

spherical particles which are dispersed in the continuous phase. The particle volume loading is usually

assumed negligible, so that particles have no feedback effect on the carrier gas and particle-particle

interactions are neglected. In the Lagrangian framework, the controlling phenomena for particle

dispersion in the field are assessed using a rigorous treatment of the forces acting on the particle. Ingeneral, the detailed flow field is computed first, then a representatively large number of particles are

injected in the field, and their trajectories determined by following individual particles until they are

removed from the gas stream or leave the computational domain. Particle motion is extracted from the

time integration of Newtons second law, in which all the relevant forces can be incorporated (drag,

gravity, lift, thermophoretic force, etc.). The Lagrangian approach is computationally intensive,

because it entails tracking a large number of particles until stationary statistics are achieved. On the

704

8/13/2019 39-Dehbi

3/18

other hand, the results of Lagrangian particle tracking are physically easier to interpret. Therefore, in

the following investigation, the Lagrangian methodology is used, along with the assumption that the

dispersed phase is dilute enough not to affect the continuous flow field (one-way coupling).

3. Particle trajectory equations

Assuming only drag is significant, the vector force balance on a particle is written as follows:

(1)

where the drag force per unit mass may be expressed as:

(2)

In the above, u is the fluid velocity, upis the particle velocity, the fluid dynamic viscosity, pthe particle density, dpthe particle geometric diameter, and Re the relative Reynolds number defined

as:

(3)

In FLUENT, the drag coefficient is obtained from the expression:

(4)

where the as are constants which apply to spherical particles for wide ranges of Re. The trajectory

L(Lx, Ly, Lz, t) of the particle is obtained by simply integrating the following vector equation with

respect to time:

dt

dLup = (5)

In laminar flow, the above relations (1) to (5) are sufficient to compute the trajectory of

individual particles, since the fluid velocity u is constant with time for steady state problems. The

particle concentration and deposition is therefore computed in a deterministic Lagrangian fashion, and

yields quite accurate results as shown by Healy [2]. In the presence of significant turbulence, the

computation of particle motion becomes considerably more complicated because of the random

velocity fluctuations which the particle faces as it is carried away by the fluid.

4. Eddy Interaction Model in FLUENT

)( pDp

uuFdt

du=

24

Re182 D

pp

DC

dF

=

ppp uud =Re

2

321

ReRe

aaaCD ++=

705

8/13/2019 39-Dehbi

4/18

A popular model to describe particle dispersion in turbulent flows is the so-called Eddy

Interaction Model (EIM) developed by Gosman and Ioannides [3] and used in FLUENT. The EIM is

a stochastic random walk treatment in which particles are made to interact with the instantaneous

velocity field u +u(t), where u is the mean velocity and u(t) the fluctuating velocity. By computingthe paths of a large enough number of particles, the effects of the fluctuating flow field can be taken

into account. In essence, the EIM aims at reconstructing the instantaneous field from the local mean

values of velocity and turbulent intensity.The EIM models the turbulent dispersion of particles as a succession of interactions between a

particle and eddies which have finite lengths and lifetimes. It is assumed that at time to, a particle with

velocity upis captured by an eddy which moves with a velocity composed of the mean fluid velocity,

augmented by a random instantaneous component which is piecewise constant in time. When the

lifetime of the eddy is over or the particle crosses the eddy, another interaction is generated with a

different eddy, and so forth. Based on the model of Gosman and Ioannides, the eddy has the following

length and lifetime:

2/34/3)(

kCLe= (6)

kC

Le= (7)

where k and are respectively the turbulent kinetic energy and dissipation rate, while Cs are

constants. In the Reynolds Averaged Navier Stokes (RANS) framework, the mean velocity u is

obtained from any of the available turbulence models (k-, k-, RSM, etc). The value of )(' tu which

prevails during the eddy lifetime is randomly drawn from a Gaussian distribution. Hence, in

component notation, one has:

2'

1' uu = (8)

2'

2' vv = (9)

2'

3' ww = (10)

where the s are random numbers with 0 mean and unit standard deviation, while the remainder of

the right-hand side represent the local rms values of the velocity fluctuations. In the RANS modeling

of turbulence, the turbulent kinetic energy kis one of the variables solved for. In the current version of

FLUENT, the fluid velocity fluctuations used for particle tracking are assumed isotropic, and the rms

values of velocity are simply obtained from the relationship:

3

22'2'2' kwvu === (11)

As already mentioned, and later confirmed in the results section, relation (11) introduces a large

over-prediction for the wall normal component v in the boundary layer, and hence a more physically

sound alternative will be developed in the next section.

706

8/13/2019 39-Dehbi

5/18

5. Inclusion of a boundary layer model in FLUENT

In the new treatment of particle turbulent dispersion, the default model with the isotropic

turbulence assumption is employed as long as the particle is outside the boundary layer, i.e. for as long

as the dimensionless particle distance to the nearest wall y+is greater than 100 (for forced flows). If

the particle crosses inside the boundary layer, then the values of the rms fluctuating velocities are

modified to account for the strong anisotropic nature of turbulence. Ideally, the rms values shouldcome from experimental or DNS data for the particular geometry and Reynolds number. In the current

implementation, DNS fits for channel flow (Re=13000) obtained by Dreeben et. al [4] are used for

both the rms velocities and turbulence dissipation rate:

496.1*

2'

)(0239.01

40.0'

+

++

+

==

y

y

u

uu (12)

421.2

2

*

2'

)(00140.0203.01

)(0116.0'

++

++

++

==

yy

y

u

vv (13)

322.1*

2'

)(0361.0119.0 +

+

+

+==yy

uww (14)

5.075.14* )(768.0)(0116.0529.4

1

)( ++

+

++==

yyu

(15)

where u* is the friction velocity and the fluid kinematic viscosity. The above represent the rms

values for the streamwise, wall normal and spanwise direction, respectively. Although these fits are

not universal, they are found to be only slightly dependant on the Reynolds number for predominantly

two dimensional flows [5], and should therefore provide better results than the default isotropic model.

The rms values of velocity are plotted in Figure 1.

In the new FLUENT model, the boundary layer modifications are made general enough so that

highly 3D flows can also be tackled. Hence, at any instant during the trajectory integration inside theboundary layer, the particle distance from the nearest wall (y+) is computed, as is the normal,

streamwise, and spanwise unit vectors at the particle location. This information allows the correct

assignment of the component fluctuating component velocities in equations (12) through (14), by first

computing these values in the local particle-wall coordinate system which varies with particle location,

then transforming theses values back to the fixed computational coordinate system.

The time integration of the trajectory equations becomes complicated very near to the wall,

because the eddy lifetime (equation (7)) tends to zero, and hence very small time steps are necessary.

A lower limit on the eddy lifetime is prescribed to avoid having to deal with extremely small times

steps. That is:

)10,

10

),7(max(7sequation

p

e

=

(16)

707

8/13/2019 39-Dehbi

6/18

1E-3 0.01 0.1 1 10 100

1E-5

1E-4

1E-3

0.01

0.1

1

+'w

+'v

+'u

Dimensionlessrmsvelocity.-

y+

Figure 1: rms values of velocity in the boundary layer

where pis the particle relaxation time representing a typical response time to changes in the carrier

fluid. It is defined as:

18

2

ppc

p

dC= (17)

where pand dpare the particle density and diameter, respectively, is the carrier fluid viscosity, and

Ccis the Cunningham correction slip factor which is very close to one for particles above 1 m (those

considered here). The integration time step is accordingly chosen to be one tenth of the eddy lifetime.

Experimentation with smaller time steps did not show any difference in the particle deposition results.

6. Results

In the following section, the assessment of the new turbulent dispersion model is given for three

different cases where experimental data on particle deposition is available: pipe flow, 90obend flow,

and a complex three dimensional internal flow. CFD Best Practice Guidelines are used to the extent

possible. In particular, the solution is determined for meshes with different resolutions to insure gridindependence, and higher order schemes are consistently used. In addition, all computations are

performed in double precision.

6.1 Pipe Flow

6.1.1 Comparison to experiments

708

8/13/2019 39-Dehbi

7/18

The deposition of particles in turbulent vertical pipe flows has been experimentally studied by a

number of authors. A good test for the new model is achieved by comparing its predictions to the well

known experiments by Liu and Agarwal [6]. The latter authors measured the deposition of 1.4 to 21

m diameter olive oil particles injected in a vertical glass pipe of diameter 1.27 cm and length 1 m.

Deposition data is presented for the portion of the pipe section between 0.25 m and 0.75 m, i.e. in the

region where the flow is fully developed and far enough from the exit. Two sets of mean gas velocitiesare used, i.e. 12 m/s and 62 m/s, which correspond to Reynolds numbers of approximately 10000 and

50000, respectively. The Liu and Agarwal correlation is stated as follows:

20006.0 ++ = pV for 13+

p (18)

1.0=+V for 13+p (19)

where V+is the particle deposition velocity scaled by the friction velocity. No uncertainty is given, but

the data is within 30% of the values given by the correlation.

In this first simulation, two sets of 3D hexahedral meshes are generated. The second, finer mesh

is obtained from the coarse mesh by simply spitting all cells except the ones adjacent to the wall into 2

in all directions, yielding 8 finer cells for any coarse cell. Because of the simple geometry and flow,the standard k-model is used, along with standard wall functions. The boundary conditions are a flat

velocity profile at the inlet and a zero gauge pressure at outlet. After convergence with 1st order

schemes is achieved, second order, then finally third order MUSCL schemes are used for all equations.

The convergence is judged by monitoring the scaled residuals along with the mean wall shear stress.

To judge convergence, FLUENT recommends a scaled residual of at least 10-3on all variables, while

simultaneously reaching stationary, iteration-independent values for some key global flow parameters

(here the wall shear stress). The computation is stopped after both of these criteria are achieved. The

summary of the simulations for both Reynolds numbers is shown in Tables 2 and 3. Convergence and

grid independence of the results are reached.

Coarse grid fine grid

Number of cells 61800 352000

Mean aspect ratio 1.1 1.1

Discretization

Scheme3rdorder MUSCL 3rdorder MUSCL

Maximum scaled

residual3.10-7 4.10-7

Mean wall y+ 30.1 30.1

Mean wall shear

stress, Pa

0.672 0.679

Table 2: Summary information for Re=10000

Coarse grid fine grid

709

8/13/2019 39-Dehbi

8/18



Discretization

Scheme

3rd

order MUSCL 3rd

order MUSCL

Maximum scaled

residual3.10

-7 6.10

-7

Mean wall y+ 135.3 129.4

Mean wall shear

stress, Pa11.91 12.00

Table 2: Summary information for Re=50000

Particle tracking is performed by releasing a large number of unit density particles which are

uniformly distributed over the flow area 0.25 m downstream of the inlet. The particles are assigned the

gas velocity at the time of release. Deposition over the next 0.5 m is monitored in order the replicatethe conditions of the test. The number of particles chosen is 10000, as preliminary simulation with

larger numbers of particles showed no change in the deposition rate beyond 5000 particles. The results

of the simulations are shown in Figures 2 and 3. Several remarks can be made: first the default model

gives totally erroneous predictions of the particle deposition rates, especially for the smaller particles.

Given the simple geometry involved, one can justifiably conclude that the default turbulent dispersion

model should not be used in wall bounded flows. On the other hand, the new model predicts

reasonably well both the trends and the magnitude of deposition. For the smallest particles, the model

over-predicts deposition rates. This feature is inherent to random walk models which produce the so-

called spurious drift [7], i.e. preferential clustering of almost massless particles in low turbulence

regions very close the walls. For fission product transport applications, particles are typically larger

and hence the model deficiency should not have significant effects.

0 10 20 30 40 50

0

10

20

30

40

50

60

70

80 Default model in

FLUENT

Pipe flow conditions:

- Pipe diameter=0.0127 m

- Deposition length=0.5 m

- Inlet velocity =12 m/s

- Re= 10000Particlefractionremoved,

%

Particle diameter, micron

Liu correlation

Model

Figure 2: Particle deposition at Re=10000

710

8/13/2019 39-Dehbi

9/18

0 5 10 15 20

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

Pipe flow conditions:

- Pipe diameter=0.0127 m

- Deposition length=0.5 m

- Inlet velocity =62 m/s

- Re= 50000Particlefractionrem

oved,

%


Liu-Agarwal Correlation

New Model in FLUENT

Default model in

FLUENT

Figure 3: Particle deposition at Re=50000

6.1.2 Effect of mesh aspect ratio

To save computational resources, high aspect ratio meshes are usually used in regions where the

flow details are unimportant, or else in the boundary layers. In the latter case, meshes with aspect

ratios of order 50-100 are commonly used, with no detrimental effect on the flow field accuracy. In

particle tracking computations, interpolation is heavily employed, because particle position rarely

coincides with the nodes. To see the effect of a moderately high aspect ratio, particle tracking

simulations were conducted on the 50000 Reynolds number case addressed earlier. As in the previous

simulations, the mesh consists of hexahedral cells, but the mean aspect ratio is 44 this time. The

summary of the simulation is given in Tables 4.

Base case mesh Stretched mesh


Mean aspect ratio 1.1 44

Discretization

Scheme3rdorder MUSCL 3rdorder MUSCL

Maximum scaled

residual3.10-7 1.10-9

Mean wall y+ 135.3 128.1

Mean wall shear

stress, Pa11.91 11.93

Table 4: Effect of high aspect ratio mesh

711

8/13/2019 39-Dehbi

10/18

As expected, the results for the stretched mesh are almost identical to those for the regular mesh,

despite a cell count which is 30 smaller.

The particle tracking results are shown in Figure 4. With the high aspect ratio mesh, the

agreement with the data is not as good as with the base case over most of the particle size range.

Because of the considerable interpolation involved in computing particle trajectories, it is best tomaintain a low aspect ratio even in boundary layer flows. One therefore should be cautious not to

consider particle tracking as a mere post-processing computation performed on a grid independent

flow field.

0 5 10 15 20

0

10

20

30

40

50

60

70

80

Re= 50000Particlefractionr

emoved,

%


Liu correlation

Aspect ratio 1.1

Aspect ratio 44

Figure 4: Effect of mesh aspect ratio

6.2 Turbulent deposition in 90obends

Studies of deposition in bends have largely concentrated on laminar flow. To the best of the

authors knowledge, the only experimental work performed in turbulent conditions is due to Pui

et al. [8] who measured the penetration of particles through bends with diameters on the order of one

centimeter. The authors provided the following correlation for the best fit through their data:

)823.2exp( == StkC

C

in

out (20)

where the Cs are the inlet and outlet aerosol concentrations, respectively, the bend angle in radians,

and Stkthe particle Stokes numbers defined as:

D

UStk

p= (21)

in which U is the mean gas velocity in the pipe, D is the pipe diameter, and pthe particle relaxation

time earlier. Again no specific uncertainty is given, but the data is enveloped by the correlation within

712

8/13/2019 39-Dehbi

11/18

a 30% band. A generic90obend section with 2 cm diameter is modeled (see Figure 5). The domain

includes a straight inlet pipe with a 5 diameter developing length (in grey), as well as an exit straight

section with 5 diameter length. A 10 m/s velocity boundary condition (Re=13000) is imposed at the

inlet, and a 0 gauge pressure is imposed at the exit.

Figure 5: A view of the 90obend section

Two hexahedral meshes are employed: a coarse mesh with 19400 cells and a fine mesh of 123000

cells. The finer mesh is obtained from the coarse mesh by halving each coarse cell direction. The cells

adjacent to the walls are preserved from this procedure to maintain the y+ at approximately the same

values for both meshes. The k- turbulence model with standard wall functions is used, since it

usually performs better thank-in curved domains. The details of the computational case are shown

in Table 5 below.

Sample velocity profiles are taken along a vertical line parallel to the Z axis (see Figure 5) which

is parallel to the exit pipe. This line cuts through the symmetry plane somewhere close to the mid

section of the bend. Profiles for the X velocity (X is parallel to the entrance pipe section) and Z

velocity are shown in Figures 6 and 7 for both meshes. From the mean values of the shear stress as

well as sample velocity profiles, one can confidently state that the computed flow is grid independent.

713

8/13/2019 39-Dehbi

12/18

Coarse grid Fine grid



Discretization

Scheme3

rdorder MUSCL 3

rdorder MUSCL

Maximum scaled

residual1.10

-9 3.10

-9

Mean wall y+ 29.1 29.1

Mean wall shear

stress, Pa 11.91 11.93

Table 5: Summary information for bend flow case

-0.045 -0.040 -0.035 -0.030 -0.025 -0.020 -0.015 -0.010

0

2

4

6

8

10

Xveloc

ity,m/s

Distance along line, m

Coarse grid

Fine grid

Figure 6: X velocity profile along a vertical line

714

8/13/2019 39-Dehbi

13/18

-0.045 -0.040 -0.035 -0.030 -0.025 -0.020 -0.015 -0.010

0

2

4

6

8

10

Zvelocity,m/s

Distance along line, m

Coarse grid

Fine grid

Figure 7: Z velocity profile along a vertical line

Particle tracking is conducted by releasing 10000 particles from the face which represents the

entrance of the bend. Particles are uniformly distributed over the face. Comparison between the

correlation and both the default and new FLUENT models is shown in Figure 8:

0 5 10 15 20 25 30

0

20

40

60

80

100

Particlefractionr

emoved,

%


Pui correlation

New modelDefault model

Figure 7: Particle deposition in 90obend

As can be seen, the agreement with the data is quite good for the new model, and acceptable forthe default model in FLUENT. The reasonable accuracy of the default model should however be

tempered by the fact that inertial impaction is dominant over turbulent impaction for most of the

particle range simulated. It can also be seen that in the lower range of particle sizes where inertia

effects decrease, the new model offers better prediction thanks to the better modeling of the turbulent

field in the boundary layer.

715

8/13/2019 39-Dehbi

14/18

6.3 Mouth Throat Geometry

One can find only scarce data relating to particle deposition in complex three dimensional

turbulent flows. Among the available data is information relative to medical aerosol removal rates in

the human extrathoracic region encompassing the mouth, the pharynx, the larynx and the trachea.

Aerosol deposition in those regions is unwarranted because the medication is meant for the lungswhich are located further downstream. Good quality, albeit limited data was recently provided by

Grigic et. al [9] who measured the removal of 3 m, 5 m and 6.5 m DEHS oil particles in an

idealized mouth-throat geometry. The latter was constructed from tomographic scans of many adult

patients with no apparent abnormalities [10] (see Figure 9). Grigic et al. used two distinct flow rates,

i.e. 30 and 90 l/min. For 30 l/min, the flow is laminar or transitional. For 90 l/min, the flow is in the

turbulent range, with a Reynolds number on the order of 8000 based on the entrance diameter of

17 mm. Therefore, the particle tracking model is compared against the 90 l/min data.

Figure 9: Mouth-throat idealized geometry

Because of the complicated 3D geometry, a full tetrahedral mesh is used. The original mesh has

1.1 millions cells with a mean aspect ratio of 1.2. The cell count is relatively high in order to be able to

resolve the boundary layer such that the low Reynolds number k-transitional flow model can be used.

For theses conditions, it is required that the wall y+ be of order 1, although higher values are

acceptable as long as they are within the viscous sublayer, i.e. y+ less than 5. A full grid-independence

study would require running a case with close to 9 million cells, clearly outside the scope of this study.

716

8/13/2019 39-Dehbi

15/18

Nonetheless, grid adaptation is performed to yield a finer grid with 2 million cells. When the solution

on the latter is converged, a further adaptation produces the finest grid with 3.3 million cells. The

particular adaptation procedure aims at providing higher cell resolution in regions with high velocity

gradients, all the while preserving the wall y+. A flat velocity profile is imposed on the inlet face of a

short 17 mm diameter pipe which provides a fully developed flow into the mouth-throat section. A 0

gauge pressure is imposed on the exit face.

Coarse grid Finer grid Finest grid

Number of cells 1.1 million 2.0 million 3.3 million

Mean aspect ratio 1.2 1.2 1.2

Discretization

Scheme2ndorder Upwind 2ndorder Upwind 2ndorder Upwind

Maximum scaled

residual6.10

-5 3.10

-4 4.10

-4

Mean wall y+ 3.72 3.57 3.57

Mean wall shear

stress, Pa 0.83 0.80 0.80

Table 6: Summary information for the mouth-throat case

A slanted line SL (see Figure 10) where the gradients are high is chosen in the symmetry plane to

look at the X and Z (vertical) velocity profiles. The x and z velocity profiles are shown in Figures 11

and 12. The computation for the coarse grid is clearly not yet grid independent. The profiles for the

finer and finest grids are nearly the same, in particular near the boundaries where it is especiallyimportant to have grid independence to ensure the accuracy of the particle deposition prediction. It

would however be beneficial to increase the grid resolution even further to resolve the slight

discrepancies which exist between the fine and finest grids in this study.

Particle tracking is performed in the finest grid with 10000 particles injected at the entrance face

of the short conditioning inlet pipe. Results for the deposition fraction are shown in Table 7. As seen

before, the default model largely over-predicts the deposition for all particle sizes. The new model

predicts deposition rates that are within 30% of the data, which is reasonable, especially in view of the

fact that the DNS fits are those of channel flows, for lack of better alternatives.

717

8/13/2019 39-Dehbi

16/18

Figure 10: Slanted line (SL) along which velocity components are displayed

-0.090 -0.085 -0.080 -0.075 -0.070 -0.065

-2

0

2

4

6

8

10

12

14

Xvelocity,m/s

Vertical Z distance, m

1.1 million cells

2.0 million cells

3.3 million cells

Figure 11: X velocity profile along slanted line SL

718

8/13/2019 39-Dehbi

17/18

-0.090 -0.085 -0.080 -0.075 -0.070 -0.065

-2

0

2

4

6

8

10

12

14

16

Zvelocity,m/s

Vertical Z distance, m

1.1 million cells

2.0 million cells

3.3 million cells

Figure 12: Z velocity profile along slanted line SL

Particle diameter, m Grgric dataFLUENT

new model

FLUENT

default model

3.0 34 5 39.0 91.0

5.0 70 3 48.2 94.8

6.5 80 3 61.4 97.8

Table 7: Deposition fraction (%) in the idealized mouth-throat geometry

7 Conclusion

A model is implemented in the FLUENT particle tracking facility in order to specifically address

boundary layer phenomena, and in particular the strong anisotropic nature of turbulence near walls.

The flow field is first obtained for 2D and 3D cases using CFD Best Practice Guidelines, in particularmultiple grid resolutions and higher order schemes. Particle tracking is then performed on the

converged, grid-independent solutions. Comparison with experimental particle deposition rates in

both 2D and 3D turbulent flows shows marked improvement over the default turbulent dispersion

model.

719

8/13/2019 39-Dehbi

18/18

8. References

[1] FLUENT, FLUENT 6.2 Users Guide (2005) Lebanon, USA.

[2] Healy, D.P., Young, J.B., 2001. Calculation of inertial particle transport using the Osiptsov

Lagrangian method. 4th Int. Conf. on Multiphase Flow, New Orleans, Paper DJ4.

[3] Gosman, A.D. and Ioannides, E., 1983. Aspects of computer simulation of liquid fuelled

combustors. J. Energy 7, pp. 482490.

[4] Dreeben, T.D., Pope, S.B., 1997. Probability density function and Reynolds-stress modeling of

near-wall turbulent flows. Phys. Fluids, 9 (1), 154-163.

[5] Antonia, R.A., Teitel, M., Kim, J., Browne, L.W.B., 1992. Low-Reynolds-number effects in a fully

developed turbulent channel flow. J. Fluid Mech. 236, 579-605.

[6] Liu, B.Y.H. and Agarwal, J.K., 1974. Experimental observation of aerosol deposition in

turbulent flow. Aerosol Sci. 5,145155.

[7] MacInnes, J.M. and Bracco, F.V., 1992. Stochastic particle dispersion modeling and the tracer

particle limit. Phys. Fluids A 4, pp. 28092824.

[8] Pui, D.Y.H, Romay-Novas, F., and Liu B.Y. H. Experimental study of particle deposition in bends

of circular cross section. Aerosol Science and Technology, 7, 301-315.

[9] Grgic, B., Finlay, W.H. and Heenan, A.F., 2004. Regional aerosol deposition and flow

measurements in an idealized mouth and throat. Journal of Aerosol Science 35, 2132.

[10] DeHaan, W. H. & Finlay, W. H. 2001. In Vitro Monodisperse Aerosol Deposition in a Mouth and

Throat with Six Different Inhalation Devices, J. Aerosol Med. 14, 361-367.

Acknowledgement

The author wishes to express his gratitude to Professor W. Finlay of the University of Alberta

for making available the CAD files of the mouth-throat geometry.

720

Documents

39-Dehbi